The Research Of Community Detection For Microblog

When online social networks are developing rapidly, MicroBlog has been recognized as a mode of new information network propagation after television, emails, newspapers, broadcasting and so on. What’s more, MicroBlog includes the features of the media and social networks. When Microblog is rising quickly, the majority of advertisers discover this opportunity on this social networking platform. Although MicroBlog social network is an only show platform to access information and knowledge and to communicate with friends for users, advertisers find this information sharing platform can have more functions, with a platform of so many registered users. Then they can use this social platform to market its products and services. Recently, these information can be passed quickly on MicroBlog platform whether it’s a text or rich media information. Thus, more and more social behavior start to spread from the offline to online, this behavior is not only reflected in the offline’s social behavior, especially, offline community’s promotion of products. Now advertisers met a problem of how to migrate from offline to online community seamlessly. Although they can use a strategy to advertise with no goal, it is no doubt that they will result in higher advertising costs. What’s more, they will lose more customers with negative advertises. If we discover related community the user belongs to, advertisers can improve the accuracy of advertising by advertising to a kind of virtual community.First, this paper presents some theoretical techniques which is related with community detection, then summarizes the current categories of community detection algorithm which contains the community detection related with label propagation, hierarchical clustering, clique and probabilistic model. Meanwhile, this paper analyzes the applied scene and merits and weakness. After we understood the idea of current categories of community detection algorithm, considering MicroBlog feature, this paper presents a parallelized multi-label propagation community detection algorithm based on the BSP model. This algorithm can solve a large-scale graph computational problems besides discovering overlapping communities that MicroBlog users belong to efficiently. The algorithm draws on the idea of the label propagation algorithm of Raghava et al, and studys the reason of LPA disadvantage, solves the weakness of the label propagation algorithm about the problem of instability and less accuracy. The algorithm solves the problem of instability by making a series of rules to choose label that the node belongs to when updating the label, rather than to choose label randomly if there are multiple optional community label. This algorithm introduces MicroBlog spreading factor to solve the problem of less accuracy, since it can identify the probability of the community that the node makes up of when the node links another node, not treating every community the node belongs to equally. The algorithm also draws on COPRA algorithm, and can discover related the structure of overlapping communities, but there are essential differences compared to CORPA:(1) The algorithm does not limit the community’s labels of each MicroBlog user belongs to, because the MicroBlog user’s interests are very broad and dynamic;(2) The algorithm discovers these active user’s related communities labels based on users tweeting, and then discovers the unlabeled user nodes according to labeled user nodes.Finally, the paper draws on a parallelized message passing mechanism based on the RDD (Resilient Distributed Dataset) model, and applies Spark GraphX Graph Computation parallelized model to the algorithm, and handles large-scale computational problems.